Embrace Complexity Through Behavioral Planning

It is no secret that simple, targeted interventions can efficiently change behavior. While there is beauty in simplicity, there are also risks when we fail to reconcile these elegant solutions with real-world complexity. When contexts rapidly shift, these solutions can become somewhat brittle or less effective than expected.

Richard Thaler recently suggested that the advancement of behavioral science may lie not in nudges or even removing “sludge” to make systems easier to navigate but in helping to craft those very systems from the outset. If behavioral science is to play an effective role in designing systems, it’s critical that we build an approach that factors in complexity.

This means expanding current norms about how behavioral designers look at challenges and propose solutions. For one, evidence-based disciplines like the behavioral sciences primarily look to the data from the past to inform solutions in the present. Drawing on rigorous data can be one of the discipline’s strengths, but, when navigating complexity, an overreliance on existing data can bias us to presume that what worked in the past will always be effective in the future. Additionally, there is no shortage of tools, frameworks, and processes for building and measuring the short-term success of interventions. But when it comes to systematically designing for complexity, there is far less guidance.

If behavioral science is to play an effective role in designing systems, it’s critical that we build an approach that factors in complexity.

Designing for complexity also requires questioning assumptions about how interventions work within systems. Being wary of three key assumptions about persistence, stability, and value can help behavioral designers recognize changes over time, complex system dynamics, and oversimplified definitions of success that may impact the effectiveness of interventions.

When behavioral designers overlook these assumptions, the solutions they recommend risk being short-sighted, nonstrategic, and destined to be reactive rather than proactive. Systematically confronting and planning for these projections, on the other hand, can help behavioral designers create and situate more resilient interventions within complex systems.

In a recent article, we explored why behavioral science is still learning to grapple with complexity, what it loses when it doesn’t, and what it could gain by doing so in a more strategic and systematic way. This approach—which we call “behavioral planning”—borrows from business strategy practices like scenario planning that play out assumptions about plausible future conditions to test how they might impact the business environment. The results are then used to inform “roughly right” directional decisions about how to move forward.

When behavioral designers overlook complexity, the solutions they recommend risk being short-sighted, nonstrategic, and destined to be reactive rather than proactive.

To help behavioral designers more systematically introduce behavioral planning when considering issues of persistence, stability, and value, we developed a framework. Its goal is to capture different factors that impact how interventions function within systems by considering two dimensions of complexity. The first dimension considers adaptation, to capture how changes at the system and the individual levels may impact interventions. The second dimension considers conditions, to capture where system changes are imposed top-down from the outside or originate more naturally from within.

In the figure below, adaptation is placed along the vertical axis and conditions along the horizontal axis. The intersections of these two dimensions prompt questions that behavioral designers can use to help address complexity when designing solutions. Below, we explore each of the four questions in more detail.

Ecosystem adaptation × human-imposed conditions: How might other policies or incentives impact interventions?

No intervention exists in a vacuum. Behavioral designers should consider the dynamics between other interventions or policies. In the Colorado River Basin, for example, a set of behavioral interventions aimed at transitioning farmers to water-saving practices was tactically successful—more farmers adopted these practices—yet it failed to decrease overall water usage. Instead, competing incentives posed by “use-it-or-lose-it” water policies at the state level washed out any gains that had been made. Designing solutions without considering broader ecosystem forces can render interventions tactical successes but strategic failures.

Behavioral designers should also be on the lookout for “substitution effects,” which occurs when adopting certain behaviors makes us less likely to follow through on others. Research has shown, for example, that nudges encouraging individuals to make higher credit-card autopayments may inadvertently prompt people to reduce the frequency or amount of manual payments required to pay off their balances. This suggests that interventions that don’t consider the full ecosystem of behavioral influences may underestimate important decision-making trade-offs.

Ecosystem adaptation × naturally emergent conditions: How are the boundaries of the problem itself changing?

Changes in the ecosystem itself can provide behavioral designers with clues for building more resilient designs. For example, saving for retirement is a long-standing and well-known behavioral challenge. However, most interventions that attempt to address it through defaults or autoenrollment are designed for traditional organizations and employer-sponsored retirement plans like 401ks. While these solutions continue to work well for these employees, they miss opportunities to address the rise in gig employment or entrepreneurial careers. People can also become desensitized to interventions. Research has shown, for example, that text messaging to encourage college matriculation has flattened, in part because of generic messaging but also because of the ubiquity of text messaging, which has diminished novelty and engagement.

Designing solutions without considering broader ecosystem forces can render interventions tactical successes but strategic failures.

Individual adaptation × naturally emergent conditions: What situational factors might impact how interventions are acted upon?

The COVID-19 vaccine rollout continues to be an enormously complex challenge that requires solving multiple simultaneous issues: vaccine hesitancy and fear of needles, tensions between individual identity and communal good, and helping ensure that people have the agency to receive the shots and recover from getting vaccinated. By more systematically considering underlying contexts, motivations, and system structures, behavioral designers can better propose and support effective solutions.

Behavioral designers must also consider the diverse, and often deeply embodied, personal experiences that inform people’s interactions with solutions. In the context of COVID-19, for example, U.S. public health recommendations to wear masks ignore Black men’s legitimate fears that donning handmade masks pits communal safety against personal safety, due to racial profiling and assumptions of criminality. Behavioral design on its own isn’t equipped to dismantle deeply embedded system inequities and social dynamics like racism. But designing for behavior change without recognizing system influences on personal choice means only solving part of the problem.

Individual adaption × human-imposed conditions: What second-order issues might we need to proactively consider?

When people adapt to a new normal, designers may need to consider the ripple effects of their original interventions. The famous example of the cobra effect—in which a mandate to reduce the cobra population through rewards for dead snakes led to a boom in snake-breeders looking to cash in—indicates the importance of considering potential downstream consequences of interventions that may prove to be a little too effective at dialing up unintended behaviors.

Considering these plausible futures can help us realize the potential of our interventions, and help address the diverse needs of people now and in the future.

Designing only with local cause and effect in mind may remove important nuances and fail to account for the emergent nature of interventions’ effect on systems, or vice versa. Consider, for example, a texting intervention by the World Bank: behavioral designers intended to encourage parenting practices in small Nicaraguan communities by sending different text messages to randomly assigned parents. However, the intervention faltered when ordinary and predictable social behaviors—like neighbors’ tendencies to chat with one another about their participation in the study, gender roles, and local power dynamics—led some communities to boycott the intervention altogether. In cases like these, taking a wider view of system dynamics and peripheral interactions may have helped behavioral designers consider the full range of inputs and outputs of potential interventions earlier on.


COVID-19 has provided no shortage of behavioral challenges to solve. The pandemic’s long arc has also highlighted the need to solve evolving challenges rather than a single challenge fixed in time. Behavioral designers should consider when to take a systems view to design interventions that address how changing contexts shape behavior, rather than focusing only on behavioral change. Initial responses to the pandemic that focused on immediate issues like working effectively from home have revealed second- and third-order effects, such as a rise in alcohol dependency, increases in domestic violence and mental health, and projected long-term losses in career advancement and earnings due to unsustainable childcare. These issues have disproportionately affected women, in particular Black and Brown women. Considering the long-term effects of interventions can, and should, impact how we design solutions. Just as importantly, they also provide insight into who is currently served by solutions and who is left out, and how we prioritize problems and audiences.

We cannot predict the future, of course, and shouldn’t try to. Even the best laid plans will never result in perfect interventions or outcomes. But becoming more alert to evolving conditions can help behavioral designers be more strategic when designing interventions within complex systems. Considering these plausible futures can help us realize the potential of our interventions, and help address the diverse needs of people now and in the future.


Disclosure: Katelyn Stenger a Ph.D. Fellow at the Convergent Behavioral Science Initiative at the University of Virginia, which provided financial support to Behavioral Scientist as an supporting partner in 2021. Supporting partners do not play a role in the editorial decisions of the magazine.